What a Voltage Sag Does to Equipment That Was Running

Why this matters

A dip lasting a tenth of a second takes out half a building's equipment and leaves the other half running, and the customer's report is "everything went down except the lights." What decides which side of that line a machine lands on is not how sensitive it is in some general sense. It is whether the voltage it retained stayed above its own dropout threshold for the whole duration of the dip, and whether it had stored energy to cover the gap if it did not. Get that model straight and you can predict, before you open anything, which devices in a panel should have survived and which should not have, and the one that did not behave as predicted is your fault.

Before you take any reading on the live side

Voltage sags are only observable on energized equipment, which puts this work squarely inside the gate at 29 CFR 1910.333(a)(1): energized work is permitted only where the employer can demonstrate that de-energizing introduces additional or increased hazards or is infeasible due to equipment design or operational limitations, and monitoring a supply to catch a transient event is the standard example of a task that qualifies. That does not make it casual work. The shock and arc flash boundaries and the PPE that goes with them come from the risk assessments at NFPA 70E-2021, 130.5 and 130.7, in the edition your employer's electrical safety program has adopted, and the meter and its leads must be rated for the system and proved live-dead-live per 120.5 before you trust anything they say. Where the panel can be recorded from a permanently installed monitoring point instead, land the recorder there and close the door.

The four events people call a sag, and how they differ

A sag (or dip) is a reduction in RMS voltage lasting from about half a cycle to a few seconds, with the supply still present.

An interruption is a loss of supply. Zero volts is a different problem, because every load loses power together and nothing is left holding a contactor in with degraded voltage.

A sustained undervoltage lasts longer than a few seconds. Different mechanism, different symptom: motors run hot rather than trip, because they carry load at reduced voltage indefinitely.

A transient is a fast overvoltage or ringing event measured in microseconds. It damages insulation and semiconductors. It is not a sag and the protection is different.

Those are four distinct events, and a monitoring instrument that logs "voltage event" without depth and duration has not told you which one you had.

Depth and duration are one gate, not two

State the gate this way, and hold to it: a device rides through a sag when the retained voltage stays above that device's own dropout or undervoltage threshold for the entire duration of the dip, OR when its stored energy covers the dip's duration. Either condition is enough for that one device. But a machine is a chain of devices in series, so the machine rides through only when every device in the chain does - one AND across the chain, one OR inside each link.

Depth is stated as percent of nominal retained, not percent lost, because that is what a device's threshold is written against. Duration is stated in cycles at the system frequency, so at 60 Hz one cycle is about 16.7 milliseconds. Use both every time. "A 30 percent sag" is not a fact until you know whether it lasted one cycle or sixty.

The design envelopes people quote from, such as the ITI curve published by the Information Technology Industry Council, plot exactly those two axes. That curve describes single-phase 120 V 60 Hz information technology equipment, it is a design reference rather than a regulation, and it binds a product only where its manufacturer declares conformity to it. It is not a promise about a contactor coil or a motor.

What each load family has in the tank

• A resistive heater has no dropout threshold at all. Power falls with the square of voltage, so at 70 percent retained it delivers 49 percent of rated power for the duration, and a hundred milliseconds of that against a thermal mass is unmeasurable.
• An AC magnet coil in a contactor or relay is the shortest fuse in most panels. Pickup commonly requires around 85 percent of rated coil voltage and dropout occurs well below that, but the actual pickup and dropout voltages are device data-sheet values. NEMA ICS 2 sits behind the common figures and binds through the manufacturer's declaration and your purchase specification, not on its own. Note also that a coil fed from a control transformer sees less than the percent retained at the panel, because the transformer's secondary tracks its primary proportionally and its own regulation drop under load subtracts further.
• A running induction motor has rotor and driven-load inertia, which is real stored energy, but its torque falls roughly with the square of terminal voltage at a given slip. At 70 percent retained that is about 49 percent of rated torque. If the driven load needs more than that, the machine decelerates for the duration of the dip.
• A variable frequency drive rides on its DC bus capacitance, and the bus falls at a rate set by capacitance and load, so ride-through is longest at light load and shortest at full load. The trip threshold and the rated ride-through are manufacturer values.
• A switch-mode power supply feeding a controller has a holdup time specified at full load, commonly in the region of 10 to 20 milliseconds, and again the number is on the data sheet.

The values used in the walkthrough below are stated as illustrative for that walkthrough, taken as if read off those data sheets, not as typical values to carry to another job.

One dip, two panels, opposite outcomes

A recorded event at the service: retained voltage 70 percent of nominal, duration 6 cycles at 60 Hz, which is 100 milliseconds. Same event reaches both panels.

Panel A runs a resistive process heater and an across-the-line pump motor held in by a mechanical contactor. The contactor's data sheet gives dropout at 65 percent of rated coil voltage. The control transformer is loaded enough that its secondary reads 66 percent of nominal when its primary is at 70 percent, so the coil sees 66 percent, one point above dropout, and the contactor holds for all 6 cycles. The heater is unaffected. The motor is the problem: at 70 percent terminal voltage it makes about 49 percent of rated torque, the pump demands more than that, and the motor decelerates for the whole 100 milliseconds because nothing disconnected it. When the supply recovers, the motor is turning below rated speed with full voltage restored, so it draws re-acceleration current somewhere between full-load and locked-rotor depending on how far it fell. That current opens the branch breaker.

Panel B runs a fan on a variable frequency drive and a controller on a switch-mode supply with 15 milliseconds of holdup at full load. At 15 milliseconds the controller's supply runs out and the controller resets. At about 50 milliseconds, roughly cycle 3 of the 6, the drive's DC bus has fallen to its undervoltage threshold and the drive trips and latches the fault. At 100 milliseconds the dip ends. The controller comes back up; the drive does not, because its fault is latched. Nothing is damaged and nothing tripped a breaker, but the fan stays off until someone acknowledges the fault.

Three events, in this order: the controller drops at 15 milliseconds, the drive trips at about 50, the dip clears at 100.

Which panel is worse off is the opposite of what the customer thinks. Panel B stopped, wrote down why, and stressed nothing. Panel A rode through in the sense the customer means, and in doing so held a motor connected to a supply that could not turn its load, then hit it with a re-acceleration it never got a decision about. The panel that reported nothing is the panel that took the damage.

What would flip this. Shorten the dip to 1 cycle, about 17 milliseconds, and Panel B's controller is marginal, the drive rides through easily, and Panel A's motor barely slows, so both panels shrug. Deepen it to 40 percent retained at the same 6 cycles and Panel A's contactor drops out, which disconnects the motor cleanly and turns Panel A into the better-behaved of the two. The two variables move the answer independently, which is why the gate needs both.

The failure mode. The breaker on Panel A gets replaced. It tests fine on the bench because there was never anything wrong with it, and the next dip trips the new one. A tech who reads the trip as evidence of a faulty device rather than as a report of a real overcurrent buys the same part twice.

Why the recovery is the half that does the damage

The instinct is that low voltage is the stress. For a load that stays connected, the stress is on the other side of the event. During the dip the motor was making about half its torque, so it slowed, and slowing raised slip, which drops its impedance toward the locked-rotor value. Current went UP, not down, at reduced voltage. On recovery it draws re-acceleration current at full voltage while turning slow, which is a locked-rotor-like condition in everything but name, and it does that with a rotor and windings that just spent the dip making heat without making torque. Where a plant has many motors on one bus, they all re-accelerate at once, and the collective re-acceleration current produces a second, deeper sag of its own that can drop out the contactors the first dip failed to drop.

That is also why sequencing matters after any event: bringing a plant back all at once reproduces the recovery transient deliberately.

Reading the aftermath against the model

• A device with a latched undervoltage fault told you the truth about what it saw, and its threshold plus the fact that it tripped bounds the depth of the event from below.
• A tripped thermal overload on a motor that has no other fault points at re-acceleration, not at the dip.
• A controller that came back with default settings or a lost clock is a holdup-time failure, not damage.
• A contactor that dropped out and did not pick back up is either a latched control scheme working as designed or a coil that never recovered enough voltage, and the difference is in whether other coils on the same transformer also stayed out.

Where the record is thin, the useful move is to put a recorder on the supply for a fortnight rather than to argue about a single event. A second event with depth and duration logged settles what a customer's memory of a flicker never will.

How to verify you got this right

Reconstruct the event from device thresholds rather than from testimony. List every device that reported a fault and every device that did not, write each one's stated threshold beside it, and check that the set of devices that tripped is consistent with a single depth-and-duration pair. If a device with a shallower threshold rode through while a device with a deeper one tripped, either the two are on different supplies, or one of them has a fault of its own, and that inconsistency is worth more than any single reading.

Then confirm the direction of the story you are about to tell. If you are going to say the motor tripped on recovery, the trip time relative to the event should be at recovery, not during the dip, and the overload or breaker that opened should be the one in the motor's own branch rather than an upstream device.

References

• 29 CFR 1910.333(a)(1) for the de-energized-work gate and (b)(2) for electrical lockout; 29 CFR 1910.147(a)(1)(ii)(C) excludes electric utilization equipment from the general lockout standard
• NFPA 70E-2021, 130.5 and 130.7 for shock and arc flash risk assessment and PPE selection, and 120.5 for live-dead-live verification, in the edition adopted by your employer's electrical safety program
• ITI (CBEMA) curve published by the Information Technology Industry Council, a voluntary design reference for single-phase 120 V 60 Hz information technology equipment
• Equipment manufacturer data sheets for coil pickup and dropout voltage, drive undervoltage threshold and ride-through, and power supply holdup time
• See related: Why Voltage Sags Under Load; Why Equipment Fails After a Power Event Rather Than During It